Stable geographically weighted Poisson regression for count data
Published Web Locationhttps://doi.org/10.25436/E2X59B
Geographically weighted Poisson regression (GWPR) is widely used for spatial regression analysis of count data. However, it tends to be unstable because of a fundamental drawback of Poisson regression. To overcome the drawback, we introduce a log-linear approximation to estimate GWPR without relying on Poisson regression framework. The proposed approach approximates GWPR using the basic GWR modeling with transformed explained variables. Monte Carlo experiments show that the proposed GWPR outperforms the conventional GWPR in terms of both estimation accuracy and computationally efficiency. Finally, the proposed GWPR is applied to an analysis of coronavirus disease 2019 (COVID-19).